TWI494900B - Method of real time image tracking - Google Patents

Description

Instant image tracking method

The invention relates to a real-time image tracking method, in particular to an instant image tracking method suitable for different actions, sizes and shapes of a target object of a template image object and an original image (to be matched).

Image tracking is currently widely used in military and video processing, and is in the field of popular development. A common image tracking method uses a full search method, which overlays the template image onto the original image and compares it with the pixel and moves the template image on the original image. Each movement is compared to all pixels. Click once and add the difference between the pixels of the two maps as the similarity of the position. When the comparison of the original image is completed, the position with the smallest sum of the differences is taken as the matching position. However, this method often takes a lot of time, so if the original image is large, it will take a very long time.

In the field of image processing, a median threshold bitmap is used to perform image calibration processing. The intermediate image of the template image and the original image is used as a threshold to convert the two images. For calibration, although this method only needs to compare the bit values of the image to process the image calibration quickly, since the image calibration is not as good as the image tracking, the template image is different from the original image. The action is not equally variable, and the median value taken when the image is heavily occupied by a certain color has a great chance for this color (especially when the image is small, that is, when the image information is small), then The median threshold bitmap of the original image and the median threshold of the template image may be different to affect the matching result, so the median threshold bitmap is not suitable for image tracking.

Therefore, it is necessary to develop a new image processing method to achieve fast image tracking.

The object of the present invention is to provide a real-time image tracking method, which is processed by an image processing device, the method comprising the steps of: (A) inputting a template image information and original image information; (B) using an image pyramid The template image and the original image are scaled to an image level of different sizes of the array, and the image levels are sequentially arranged according to the image size; (C) the threshold of the template image and the original image in the minimum image level is performed. a map conversion for respectively obtaining an average threshold map of the template image and the original image; (D) performing a similarity matching between the template image and the average image threshold of the original image, thereby finding out The template image is matched to one of the original image images; (E) the image matching position obtained in the step (D) is mapped to the original image image of the next image level, and the range is expanded by the matching position, and the image is Converting the template image of the image level to the threshold image of the original image, and then only matching the template image with the average threshold of the original image for the range; and F) repeating step (E) until the image level is the original size level of the images, and the matching of the levels is Output as a match result. In this way, objects matching the template image can be found in the original image. In addition, the present invention can also add an enlarged original image as the last layer of the image pyramid, and then perform a template matching process, and finally remove the matched bits by two, thereby obtaining a more accurate matching position.

11, 31‧‧‧ original image

511‧‧‧Minimum image level matching position

12, 32‧‧‧ template image

52‧‧‧Next level image level

13‧‧‧Image processing device

521‧‧‧Next image level matching position

14‧‧‧Complete completion diagram

53‧‧‧ Original level image level

S21, S22, S23, S24, S25, S26, S27, S28‧‧‧ steps

531‧‧‧Final matching position

S31, S32, S33, S34‧‧‧ steps

S61, S62, S63‧‧‧ steps

33‧‧‧Average threshold image of the original image

S71, S72, S73, S74, S75, S76, S77, S78‧‧

34‧‧‧Average threshold map of template images

S81, S82, S83, S84, S85, S86, S87, S88‧‧

S41, S42, S43, S44, S45‧‧ steps

S91, S92, S93‧‧‧ steps

50‧‧‧Image Pyramid

101‧‧‧Enlarge the original pixel in the image

51‧‧‧Minimum image level

102‧‧‧First type interpolation pixel

103‧‧‧Second type interpolated pixel

Figure 1 is a schematic illustration of image matching of the present invention.

2 is a flow chart of the image matching method of the present invention.

3 is a flow chart showing the conversion of the average threshold map of the present invention.

4 is a flow chart showing the matching of the average threshold map of the present invention.

Figure 5 is a flow chart of the image pyramid of the present invention.

6 is a flow chart of the present invention in conjunction with a mutually exclusive or bit map operation.

Fig. 7 is another flow chart of the image matching method of the present invention.

FIG. 8 is a flow chart of the method for obtaining a more accurate matching position according to the present invention.

FIG. 9 is a flow chart of another method for improving matching accuracy according to the present invention.

Figure 10 is a schematic illustration of the bilinear interpolation method of the present invention.

The present invention provides an instant image tracking method. Referring to FIG. 1, it can be seen that the tracking intention is to find an object (14) matching the template image (12) on the original image (11), for example, "C shown in FIG. One of the images can be considered to be composed of a plurality of pixels (121) arranged in a matrix. For example, the template image (12) shown in FIG. 1 is actually composed of a plurality of pixels (121). The method is implemented by an image processing device (13), and the image processing device can be A computer, a microprocessor, etc., as long as it is any device having a computing function, can be used as the image processing device of the present invention.

2 is a flowchart of an embodiment of the present invention. First, in step S21, an image processing apparatus obtains a template image and an original image to be matched. Then, in step S22, the image processing device uses an image pyramid to scale the template image and the original image into an array of images of different sizes, and divides the image into several image levels, and proceeds to step S23 to image the images according to the image. The size is arranged in order, and the ordering ranks the smallest level at the top of the image pyramid as the first level, and the original image size is arranged at the bottom of the image pyramid, wherein the image size of each level can be an integer The multi-scale, for example, the difference of each level is twice the size, but is not limited thereto. After the levels are arranged in sequence, step S24 is performed to perform an average threshold map conversion between the template image and the original image in the minimum image level to obtain the average threshold of the template image and the original image. The meta-graph, the average threshold map will be explained in more detail in the following paragraphs.

After obtaining the average threshold value map of the two images, proceeding to step S25 to perform similarity matching on the template image and the average threshold map of the original image, thereby finding out that the template image is in the original One of the matching locations in the image and the matching center block of the matching location will be described in more detail in subsequent paragraphs. Then, in step S26, the matching center block is corresponding to the original image of the next level by the matching center block obtained in step S25, because the original image of the next level must be larger than the original level of the previous level. The image is imaged, so the corresponding matching center block needs to be expanded outward to a range. Then proceeding to step S27, Performing an average threshold map conversion between the template image and the original image in the next level to obtain an average threshold map of the template image and the original image in the next image level, and then performing template in the range The image is matched with the similarity of the average threshold of the original image. The average threshold of the original image is preferably converted to capture only the image in the range, but can also be integrated. The original image is converted or the range to be converted can be selected. Then, in step S28, it is determined whether the image size is the original image size. If it is not the original image size, steps S26 to S27 are repeated to perform the next level of similarity matching. If the original image size has been reached, the matching position of the level is completed. Output, the matching position is regarded as the object of the template image on the original image, thereby completing the tracking of the template image on the original image.

3 is a detailed flow chart of the average threshold 槛 bitmap conversion of step S24 or S27. First, step S31 is performed to perform a color conversion between the original image (31) and the template image (32), and the color conversion is preferably gray. The order conversion is not limited thereto, and for clarity of explanation, the gray scale conversion will be exemplified below. After the gradation conversion, step S32 is followed to set an average value as the threshold value of the subsequent bitmap conversion, and the average value may be an average value of the grayscale values of the template image and the original image respectively (ie, the template image) Using the average of the template image, the original image uses the average of the original image), but it is better that the template image and the original image use the average value of the template image grayscale value as the average of the threshold image operation, the template The advantage of using the template image grayscale value as the average value between the image and the original image is that when the average value is calculated, the average value of the original image and the template image may be greatly different (for example, because the original image is backed by the image). The image is too complicated, and the original image will generally contain the object of the template image, so the template image as the average will achieve the most accurate calculation results. After the average value is set, steps S33 and S34 are performed, and the average threshold value map of the template image and the original image is obtained by using the following conditions: Where x is the grayscale value of the image and Mean is the average of the grayscale values of the image. After the value is converted by the average threshold, if the pixel on the template image and the original image is larger than the average value (ie, 1), it will be displayed in white. If it is less than or equal to the average value (ie, 0), it will be displayed in black, but not It is limited to white and black, as long as it is two different colors and can be used as the converted display color. The converted original image (33) and template image (34) are then output.

4(a) is a schematic diagram of the similarity matching of the present invention. First, step S41 is performed to overlay the average bitmap of the template image on the average bitmap of the original image, and step S42 is performed in the template image. Performing a point-to-point pixel value comparison operation of two images in the range of the image size and the size of the template image, and obtaining a comparison result, and then performing step S43 to perform an operation on the comparison result, which is preferably mutually exclusive. Or (XOR) operation, but is not limited thereto, and for the sake of clarity, the following will be a mutually exclusive or arithmetic operation. 1 is output when 0 and 1 are input at the same time of mutual exclusion or operation. Otherwise, it is 0. Therefore, when comparing, the average value of the template image and the average bitmap of the original image are pixel points. When they are the same, a value of 0 will be obtained after the mutual exclusion or operation, and when the pixel points are not the same, a value of 1 will be obtained. Then, in step S44, all the values in the template image range are added, that is, the difference degree value of the template image at the position of the original image. The average bitmap of the template image is then moved to the next position, and the similarity is calculated, and so on. After the template image has moved all the image range of the original image, step S45 is performed to compare the degree of dissimilarity of each position, wherein the position of the difference is the smallest (the least one obtained) is the template image on the original image. The location that is matched. Figure 4 (b) is a matching diagram, please refer to Figure 3 together, wherein the schematic 41 is a template image (34) in the original image (33) in the vicinity of the template image object position (411) for pixel comparison process, due to The pixels (black and white) are not exactly the same. Therefore, assuming that there are 60 pixels in which the difference is common, the 60 points are recorded as 1, and the dissimilarity value 60 is obtained. The schematic image of the 42-module image (34) is compared with the pixel at another position of the original image (33), since the position (412) in the original image (34) is converted to a black pixel by the bitmap. Therefore, the difference degree value will be higher than the alignment position of the schematic diagram 41. If there are 100 pixels of the difference, the 100 points are recorded as 1, and the difference degree value 100 is obtained. The diagram 43 is a template image (34) for pixel comparison at the template object position (413) of the original image (33). Since the white pixels and the black pixels are closest to each other, the difference value will be the smallest. , so set this to the matching location.

FIG. 5 is a schematic diagram of an image pyramid of an original image. Referring to FIG. 1 together, when the minimum image level (51) is matched, a matching position of the template image of the minimum image level in the original image is obtained ( 511), since the image pyramid system gradually enlarges the image from the layer of the minimum image, the matching position (511) can be used as the next image level (52) The reference of the Central Plains image, but because at the small image level, because the image reduction loses most of the image information, it is necessary to perform matching processing for all the original images at the minimum image level. However, since the size of the object to be matched (521) in the next image level (52) is also enlarged as the image is enlarged, the object (511) included in the matching position is smaller, and the matching position is also It is possible to slightly shift, so the obtained matching range (511) must be expanded. The expansion is outwardly expanded by the matching position as the center of the expansion. In a preferred embodiment, the expansion is a 3×3 ratio (ie, 9 times the size), but is not limited thereto, thereby ensuring a matching position range. Precision.

Therefore, after the matching position (511) is found in the minimum image level by the image pyramid, the corresponding matching position (512) can be quickly found in the subsequent larger image level according to the position, and only in the This level of matching is done around the location until all levels have been matched, ie when the level is the original image size (53), the actual match is completed.

Thereby, the present invention can provide a fast image tracking method.

In addition, in the zooming of each image level, some image levels may be scaled very small, which may cause insufficient information of the template image. If only the template image with insufficient image information is matched to the current matching position obtained at the previous level, Searching at the matching range of the level may cause the matching position to be incorrect, because the images are not complete, so when the image information of the current level is insufficient, the search range of the upper level may not be used for searching. A full search of the original image is performed using the full search method. The image information can be various information, preferably a template image. The number of pixels is not limited thereto, and in order to make the description more detailed, the number of pixels will be exemplified below. In addition, the determination of whether the image information is sufficient or not is preferably set such that when the number of pixels of the template image exceeds 512, the image information is deemed to be sufficient.

In addition, the steps of the average threshold map can be combined with other bitmap operations. Since the average threshold is used, in the region where the light and shadow are interlaced, the pixel values of some light spots are often close to the average value, so the average threshold pixel map may be caused by the light spot noise, so the present invention can In combination with the mutex or bit map, the spot noise is filtered to obtain a more accurate average threshold map. FIG. 6 is a flowchart of a mutual exclusion or operation according to the present invention. Referring to FIG. 2 and FIG. 3 together, step S61 is first performed to set a noise threshold range, which is based on the average of the template images. The value ±r is used as the noise threshold range. Then, in step S62, the pixel values of the template image and the original image in the range (ie, the spot noise) are recorded as 0, and the template images and the pixels of the original image are not in the range. The value (ie, non-spot noise) is recorded as 1, thereby obtaining the mutual exclusion or bit map, wherein the r value depends on the difference between the pixel value of the spot noise and the average value, such as spot noise. The value is 172, and the average is 168, then the r value is 4. Then, in step S63, the average threshold value bitmap obtained by the flow of FIG. 2 is ANDed with the mutual exclusion or bit map, and the portion where the pixel values of both the graphs are all 1 will be 1 after the AND operation. Output, so the mutex noise (ie 0) on the mutex or bit map will be filtered out, so the image after the AND operation will be a more accurate average threshold map, ie the light point is removed. The average threshold value of the noise.

Thereby, the present invention can provide a fast and accurate image tracking method.

FIG. 7 is a flowchart of another embodiment of the present invention, wherein the flow is different from the flow of the embodiment of FIG. 2 in that the flow of FIG. 2 is to first scale the original image and the template image into several classes. The conversion of the average threshold map is performed when the class is performed. In this embodiment, the conversion of the original image of the original image and the average threshold of the template image is performed before the image is scaled, and the conversion is performed. The resulting image is scaled. The detailed process of this embodiment is as follows. First, step S71 is performed to input a template image and an original image to be matched to an image processing apparatus. After inputting the image, proceeding to step S72, performing an average threshold map conversion between the template image and the original image to obtain an average threshold value map of the template image and the original image, the average threshold The acquisition of the meta-graph is the same as in the previous embodiment. Then, in step S73, the template image and the average threshold of the original image are scaled into an array of different size images by using an image pyramid, and then divided into several image levels, and then step S74 is performed. The image hierarchy is sequentially arranged according to the size of the image. The sequential arrangement ranks the smallest level at the top of the image pyramid as the first level, and the level of the original image size is arranged at the bottom of the image pyramid, wherein each original image The size may be an integer multiple of the scaling, for example, the difference of each level is 2 times the size, but is not limited thereto.

After the levels are arranged in sequence, step S75 is performed to perform similarity matching between the template image and the average threshold image of the original image, thereby finding a matching position of the template image in the original image and match The center pixel block is the same as the previous embodiment. Then, in step S76, the matching center pixel block of the matching position is obtained in step S75, and the matching center block is corresponding to the original image in the current level, because the original image of the current level is larger than the original image of the upper level. Therefore, the corresponding matching center block is outwardly expanded to a range, and the similarity matching between the template image of the current level and the average threshold of the original image is performed in the range. Then, in step S77, it is determined whether the image size of the current level is the original image size, and the similarity matching of the next level is performed. If the original image size is used, the matching position is output, and the matching of the template image on the original image is completed.

In addition, FIG. 8 is another embodiment of the present invention, which differs from the embodiment of FIG. 7 only in that the flow of FIG. 7 is changed to the zooming of the image pyramid of step S82 and the image of the image by step S83. Arrange, and in step S84, the image of each level is simultaneously converted into a bitmap, and then the image matching of each level of steps S86 to S88 is performed.

The present invention also proposes a method for improving the matching accuracy. As shown in FIG. 9, after the matching of the original image size layer of the image pyramid in the above embodiment is completed, step S91 may be further performed to image the original image of the original image layer and the template. The image is doubled by a bilinear interpolation method, and then step S92 is performed, and the enlarged original image is used as the last layer of the image pyramid, and matched with the template image, after the matching position is obtained, Step S93 is performed to divide the enlarged matching position coordinate value by 2, thereby obtaining a more accurate matching position. The image to be enlarged may be an image of an original image size layer that has been subjected to bit value conversion, or Therefore, the image has not been converted by the bit value, that is, the bit value conversion can be performed after the enlargement, and after the enlarged original image is added to the image pyramid, the matching position of the original image level can also be obtained as in the foregoing embodiment. Mapping to the magnified image level forms a range, after which only matching is performed within the range, thereby reducing the time of matching.

10 is a schematic diagram of a bilinear interpolation method, in which a pixel having a plurality of original images (101) and a first type of interpolated pixel between each original image pixel (101) are imaged in a magnified image. Point (102) and a second type of interpolated pixel point (103), wherein the first type of interpolated pixel point (102) is a pixel point between two original image pixel points (101), the pixel value of which is An average of the two original image pixel values, the second type of interpolated pixel point (103) is a pixel point between the four original image pixel points (101), and the pixel value is the four original image pixel values The average value. Since the magnified image has more pixel values, the match between the original image and the template image will be more accurate. Since the position coordinate value on the enlarged image is also twice the value of the original size image coordinate, after processing the matching of the enlarged image, the coordinate of the obtained matching position is divided by 2 to obtain the original size image. The coordinates of the coordinates. Moreover, zooming in twice is only a preferred embodiment, and the present invention can also amplify other multiples.

In the above embodiments, the average threshold pixel map can be combined with the mutual exclusion or bit map to obtain a more accurate average threshold map.

In the above embodiment, the full search method can be used for matching matching on the image pyramid level where the image information is insufficient.

Therefore, the present invention can provide a fast image tracking method, and Use the average threshold to filter the image information and use the image pyramid to scale the image and divide it into several levels, and then use the matching position information of the images between the levels to reduce the search range of the next level, and can be used when the image information is insufficient. Significantly reduce the time for image tracking. In addition, the present invention utilizes the average value of the pixel values of the template image as the threshold of the bitmap, and obtains a precise bitmap by mutual exclusion or operation, and then interleaves the bitmap of the template image with the bitmap of the original image. Repel or calculate, find a range closest to the template image on the original image as a match, whereby the pixel value of the template image object does not have much difference in different actions, angles, etc., so the present invention Applicable to the template image and the object to be matched in the original image are exactly the same, the same object but different actions, the same object but different background, the same object but different angles or the like or a combination of these different situations. In addition, the present invention can further add an enlarged image matching step to obtain a more accurate matching position.

The above-mentioned embodiments are merely examples for convenience of description, and the scope of the claims is intended to be limited to the above embodiments.

S21, S22, S23, S24, S25, S26, S27, S28‧‧‧ steps

Claims (30)

An instant image tracking method is processed by an image processing device, the method comprising the steps of: (A) obtaining a template image information and original image information; (B) using an image pyramid to image the template image with an original image The image is scaled to an image level of different sizes of the array, and the image levels are sequentially arranged according to the image size; (C) the threshold image conversion of the template image and the original image in the minimum image level is performed to obtain the image An average threshold value map of the template image and the original image; (D) matching the template image with the average threshold value of the original image, thereby finding the template image in the original image a matching position; (E) correspondingly matching the matching position obtained in the step (D) to the original image image of the next image level, and expanding a range from the matching position to perform the template image and the original image of the image level The threshold of the image is converted by the bitmap, and then only the template image is matched with the average threshold of the original image for the range; and (F) the step (E) is repeated until the shadow Level for such an image size of the original level; whereby to match the position of the last stage outputs the result as matching. The method for tracking an instant image according to claim 1, wherein the step (F) further comprises a step (G) for transmitting the original image and the template image of the original size image layer by a bilinear interpolation. Method (Bilinear Interpolation) enlarges an integer multiple to be the last level of the image pyramid, and performs matching of the enlarged original image with the template image, and then divides the coordinate value of the matching position by the magnification to obtain a more accurate Matching location. The instant image tracking method of claim 2, wherein the average of the average threshold map is an average of pixel values of the template image. The method for tracking an instant image according to claim 2, wherein the similarity matching of the step (D) is mutually exclusive or (XOR) the pixel of the template image and the average threshold of the original image. ) Operation. The instant image tracking method of claim 2, wherein the range of the step (E) is outwardly expanding centering on the matching position. The instant image tracking method of claim 2, wherein when the image information of the hierarchy is insufficient, the search is performed by a full search method (Full Search). The method for tracking an instant image according to claim 2, wherein the obtaining of the average threshold map of the step (C) or (E) further comprises an average threshold of the template image and the original image. The bit maps are respectively subjected to an AND operation with the mutual exclusion or bit map of the template image and the original image to obtain an accurate average threshold map of the template image and the original image, and The accurate average threshold map is used to replace the original average threshold map for the subsequent steps. The method for tracking an instant image according to claim 7, wherein the bitwise AND operation is to mutually exclusive or bit map interlaced regions The impurity point on the upper side is set to 0, and after the operation is performed, the impurities are eliminated, wherein the determination of the impurity point determines whether the value of the pixel point is within the impurity point range, and if it is within the impurity point range, it is judged The pixel points are impurity points. For example, the instant image tracking method described in claim 2, wherein the image size difference of each level of the image pyramid is an integer multiple. The instant image tracking method of claim 1, wherein the template image and the object to be matched in the original image may be identical, the same object but different actions, the same object but different backgrounds, the same object but Situations such as different angles or combinations of such different situations. An instant image tracking method is processed by an image processing device, the method comprising the steps of: (A) obtaining a template image information and original image information; (B) using an image pyramid to image the template image with an original image The image is scaled to an image level of different sizes of the array, and the image levels are sequentially arranged according to the image size; (C) the threshold image conversion of the template image and the original image image in all image levels is performed to obtain the image (1) performing a minimum threshold of the template image and the original image; (D) performing a minimum image layer concentration matching the template image with the average threshold value of the original image, thereby finding the template image One of the original image matches the position; (E) the matching position obtained in step (D) is mapped to the original image of the next image level, and a range is expanded by the matching position, and then only the range is Performing the average threshold of the template image and the original image Metagraph similarity matching; and (F) repeating step (E) until the image level is the original size level of the images; thereby outputting the matching position of the last level as a matching result. The method for tracking an instant image according to claim 11, wherein the step (F) further comprises a step (G) for transmitting the original image and the template image of the original size image layer by a bilinear interpolation. The method enlarges an integer multiple to be the last level of the image pyramid, and performs matching between the enlarged original image and the template image, and then divides the coordinate value of the matching position by the magnification, thereby obtaining a more accurate Match the location. The instant image tracking method of claim 12, wherein the average of the average threshold map is an average of pixel values of the template image. The method for tracking an instant image according to claim 12, wherein the similarity matching of the step (D) is mutually exclusive or (XOR) the pixel of the template image and the average threshold of the original image. ) Operation. The instant image tracking method of claim 12, wherein the range of the step (E) is outward expansion centering on the matching position. The instant image tracking method of claim 12, wherein when the image information of the hierarchy is insufficient, the search is performed by a full search method (Full Search). The method for tracking an instant image according to claim 12, wherein the average threshold of the step (C) or (E) is obtained. The step includes performing an AND operation between the template image and the average threshold of the original image and the mutual exclusion or bitmap of the template image and the original image to obtain the template image and the original image. The accurate average threshold image of the image is taken, and the original average threshold map is replaced by the original average threshold map to perform the subsequent steps. The real-time image tracking method according to claim 17, wherein the bitwise AND operation is performed to set the impurity point on the mutually exclusive or bit map interlaced area to 0, thereby performing the operation. The impurities are eliminated, wherein the determination of the impurity point determines whether the value of the pixel is within the range of the impurity point, and if it is within the impurity point, the pixel point is determined to be an impurity point. For example, the instant image tracking method described in claim 12, wherein the image size difference of each level of the image pyramid is an integer multiple. The method for tracking an instant image according to claim 12, wherein the template image and the object to be matched in the original image may be identical, the same object but different actions, the same object but different backgrounds, the same object but Situations such as different angles or combinations of such different situations. An instant image tracking method is processed by an image processing device, the method comprising the steps of: (A) obtaining a template image information and an original image information; and (B) performing an average of the template image and the original image. Threshold bitmap conversion; (C) using an image pyramid to scale the average threshold of the template image and the original image to an image hierarchy of different sizes of the array, the image levels are sequentially arranged according to the image size; (D) performing a minimum image layer concentration matching the template image with the average threshold value of the original image, thereby finding a matching position of the template image in the original image; (E) step (D) The obtained matching position corresponds to the average threshold 槛 bitmap of the original image image of the next image level, and a range is expanded by the matching position, and then the template image and the original image are only used for the range. The average threshold map is similarly matched; and (G) repeats step (E) until the image level is the original size level of the images; thereby, the matching position of the last level is output as a matching result. The method for tracking an instant image according to claim 21, wherein the step (G) further comprises a step (H) for transmitting the original image and the template image of the original image size level through a bilinear interpolation. The method enlarges an integer multiple to be the last level of the image pyramid, and performs matching between the enlarged original image and the template image, and then divides the coordinate value of the matching position by the magnification, thereby obtaining a more accurate Match the location. The method for tracking an instant image according to claim 22, wherein the average of the average threshold map is an average of pixel values of the template image. The method for tracking an instant image according to claim 22, wherein the similarity matching of the step (D) is mutually exclusive or (XOR) the pixel of the template image and the average threshold of the original image. ) Operation. The instant image tracking method of claim 22, wherein the range of the step (E) is outward expansion centering on the matching position. The instant image tracking method of claim 22, wherein when the image information of the hierarchy is insufficient, the search is performed by a full search method (Full Search). The method for tracking an instant image according to claim 22, wherein the obtaining of the average threshold map of the step (C) or (E) further comprises an average threshold of the template image and the original image. The bit maps are respectively subjected to an AND operation with the mutual exclusion or bit map of the template image and the original image to obtain an accurate average threshold map of the template image and the original image, and The accurate average threshold map is used to replace the original average threshold map for the subsequent steps. The real-time image tracking method according to claim 27, wherein the bitwise AND operation is performed to set the impurity point on the mutually exclusive or bit map interlaced area to 0, thereby performing the operation. The impurities are eliminated, wherein the determination of the impurity point determines whether the value of the pixel is within the range of the impurity point, and if it is within the impurity point, the pixel point is determined to be an impurity point. For example, the instant image tracking method described in claim 22, wherein the image size difference of each level of the image pyramid is an integer multiple. The method for tracking an instant image according to claim 22, wherein the template image and the object to be matched in the original image may be identical, the same object but different actions, the same object but different backgrounds, the same object but Situations such as different angles or combinations of such different situations.